On the Evaluation of Stress Triaxiality Fields in a

Notched Titanium Alloy Sample via Integrated DIC

Dominik Lindner, Florent Mathieu, François Hild, Olivier Allix, Cuong Ha
Minh, Olivier Paulien-Camy

To cite this version:

Dominik Lindner, Florent Mathieu, François Hild, Olivier Allix, Cuong Ha Minh, et al.. On the
Evaluation of Stress Triaxiality Fields in a Notched Titanium Alloy Sample via Integrated DIC.
Journal of Applied Mechanics, American Society of Mechanical Engineers, 2015, 82 (7), pp.4030457.
�10.1115/1.4030457�. �hal-01166848�

HAL Id: hal-01166848

https://hal.archives-ouvertes.fr/hal-01166848
Submitted on 23 Jun 2015

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci-
entific research documents, whether they are pub-
lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
 On the Evaluation of Stress Triaxiality Fields in a
Notched Titanium Alloy Sample via
Integrated DIC

Dominik LINDNER∗, Florent MATHIEU, François HILD, Olivier ALLIX†, Cuong HA MINH
LMT, ENS Cachan/CNRS/Université Paris Saclay
61 avenue du Président Wilson, F-94235 Cachan Cedex, France
Email: allix@lmt.ens-cachan.fr

Olivier PAULIEN-CAMY
Turbomeca (SAFRAN Group), avenue Joseph Szydlowski, F-64511 Bordes, France
Email: olivier.paulien@turbomeca.fr

This paper presents a coupled experimental/numerical pro-
cedure to evaluate triaxiality fields. Such a type of analysis
is applied to a tensile test on a thin notched sample made
of Ti 6-4 alloy. The experimental data consist of digital im-
ages and corresponding load levels, and a commercial code
(Abaqus) is used in an integrated approach to Digital Image
Correlation (DIC). With the proposed procedure, samples
with complex shapes can be analyzed independently with-
out having to resort to other tests to calibrate the material
parameters of a given constitutive law to evaluate triaxili-
ties. The regularization involved in the integrated DIC pro-
cedure allows the user to deal with experimental imperfec-
tions such as cracking of the paint and/or poor quality of the
speckle pattern. For the studied material different hardening
postulates are tested up to a level of equivalent plastic strain
about three times higher than those achievable in a tensile test
on smooth samples. Different Finite Element discretizations
and model hypotheses (i.e., 2D plane stress and 3D simula-
tions) are compared.
1 Introduction
In the design of turbo-engines, it is required to ensure
safety margins between the operating conditions and burst.
A criterion, which is based upon the calculation of the aver-
age hoop stress, was proposed by Robinson [1] and is used
to estimate the burst speed of disks. In the case of complex
geometries and material models, this criterion is not accurate
enough. Today’s trend is to try to reduce safety margins by
relying on local burst predictions where finite element simu-
lations are performed to take into account precisely both the
complex geometry of actual disks and the constitutive model
of the material [2, 3].
∗ also at Turbomeca
† Corresponding author
The burst prediction through local approaches strongly
depends on the modeling and identification of the constitu-
tive behavior [2, 3]. For the studied material, a Hosford yield
criterion [4] was found to be more appropriate than the J2-
flow rule. However, even if the burst predictions were im-
proved in comparison with those based on Robinson’s cri-
terion, the remaining discrepancy between experiments and
numerical predictions has led the aero-engine manufactur-
ers to look for possible explanations and better predictions.
A possible source of discrepancy is that the state of stress
in the disks, which is estimated via 3D simulations with
von Mises’ elastoplastic models identified on uniaxial tensile
tests on smooth samples (i.e., with triaxialities of the order of
0.33), is characterized by higher levels of stress triaxiality for
defect-free configurations (i.e., ranging from 0.5 to 0.8).
The influence of the stress triaxiality [5, 6] and even
Lode parameter [7, 8] on the plastic behavior of metal and
alloys has been the subject of many studies [9–14]. More-
over the stress triaxiality is a key parameter when dealing
with damage and failure [15–19]. In addition to the depen-
dence on triaxiality, the studied titanium alloy exhibits an
important tension-compression asymmetry, which requires
the use of enhanced plasticity models [20]. For the appli-
cations of interest, the triaxiality is always positive, which
allowed us to test simpler material models. It was decided
to conduct experiments on flat notched specimens giving ac-
cess to levels of triaxiality consistent with the simulations of
defect-free disks and allowing for the use of digital image
correlation (DIC) techniques. Such sample geometries in-
duce nonuniform strain fields. In that case the measurements
of load-dependent displacement fields and their subsequent
exploitation to derive strain fields are needed [21,22]. Differ-
ent optical techniques can be used to measure displacement
fields [23, 24]. Among them, DIC is increasingly employed
due to its versatility and applicability to “any” scale of obser-
vation [25, 26]. Most of nowadays DIC codes rely on local least the lower bound of the targeted triaxiality range (i.e.,
registrations (i.e., with subimages [27]). Global approaches 0.50).
provide an alternative, which uses kinematic bases that are
made consistent with numerical simulations (e.g., finite ele-
ment DIC [22, 28, 29]).
Even though very powerful, finite element DIC as any
DIC approach suffers from the ‘resolution/spatial resolution’
curse since it is an inverse problem [30]. In particular, it does
not allow FE meshes of arbitrarily small sizes to be used.
For the test that is analyzed herein this limitation occurs for
minimal element sizes of 60 pixels due to the poor random
texture. For such element sizes the mechanical FE solution
is not refined enough. To correct for such shortcomings, a
further step is to integrate even more closely measurements
and simulations. This is achieved by resorting to Integrated
DIC (I-DIC) that uses mechanically admissible displacement
fields as kinematic bases. Such a type of technique was first
applied with closed-form (i.e., elastic) solutions [31,32], and Fig. 1. Triaxiality field obtained by 3D FE elastoplastic simulations
then generalized to numerically generated solutions [33–35]. on the final sample geometry. Three midsection cuts are shown
The FE code is used, non intrusively, as a kinematic basis
generator and virtual load cell (via sensitivity fields) and the
unknown degrees of freedom become either boundary con- The chosen notched sample has a ligament width of
ditions and material parameters [35]. 3 mm, a notch radius of 1.5 mm, and a thickness equal to
The present paper shows a combined use of DIC and I- 0.7 mm, see Figure 2). It is made of Ti 6-4 alloy.
DIC to evaluate the in situ stress triaxiality in a thin notched
sample made of titanium alloy. With the proposed procedure,
each sample can be analyzed independently without having
to resort to other tests to calibrate the material parameters
of a given constitutive law. Moreover the regularization em-
bedded in the I-DIC procedure allows us to deal with exper-
imental imperfection such as cracking of the paint and poor
quality of the speckle pattern. In the first part the choice of
the sample geometry and the experimental procedure are de-
scribed. The principles of DIC and I-DIC are briefly summa-
rized in the second part. This technique gives access to the
measured boundary conditions that can be prescribed when
performing an identification step either using Finite Element
Model Updating [36, 37] (FEMU) or I-DIC [35]. The latter
is utilized herein with different constitutive laws. The results Fig. 2. Sample geometry (nominal dimensions in mm)
obtained via DIC and I-DIC are analyzed in the third part.
Various configurations are studied, namely, 2D and 3D FE
simulations, discretizations and constitutive equations. Last, The region in the vicinity of the notches has been mon-
the triaxiality fields are assessed. itored with a single camera (definition: 1388 × 1038 pixels,
16-bit digitization) and a telecentric lens to minimize the ef-
fects associated with out-of-plane motions. With the chosen
2 Experimental Setup optical setup, the physical size of one pixel is 6.4 µm so that
A drawback of uniaxial tensile tests on smooth samples the region of interest (ROI) encompasses the thinned regions
is that the onset of localized strain fields occurs at low strain (Figure 4). Since the sample surface had to be polished to
levels making it ill-suited for constitutive and failure char- reach the desired thickness, the texture was more difficult to
acterization [38]. Furthermore the state of stress at failure deposit. Two series of black and white droplets were sprayed
may be different from that of in-service structures. In the and resulted in a rather coarse speckle pattern, which is very
present study the state of stress is characterized by its von- challenging for standard DIC approaches. The experiment
Mises level and corresponding triaxiality. has been performed in a displacement-controlled mode to al-
FE simulations are used to design the sample geometry low for softening.
(Figure 1). They are carried out up to 0.12 cumulated plastic In Figure 3 the influence of the observation scale on the
strain levels, which is the macroscopic strain to failure for strain response of the specimen is presented. The net sec-
the chosen material tested in tension on a smooth geometry tion stress corresponds to the applied load F divided by the
and measured with an extensometer. The aim is to reach at initial surface S0 of the ligament. From the Q4-DIC results
to be presented in the next section, the macroscopic longi-
tudinal strain corresponds to the mean strain over the whole
region of interest. This information would have been avail-
able if an extensometer had been used with a gage length
L0 = 3.97 mm. The corresponding mesoscopic strain cor-
responds to the average longitudinal strain in an area center
about the ligament for a length L0 = 0.37 mm. Because of the
stress and strain concentrations, the mesoscopic strain level
is significantly higher than that at the macroscopic level. In
the following, all results will be related to a loading parame-
ter here chosen to be macroscopic strain εmacro .
1,200
1,150
1,100
Net section stress, F/S0 (MPa)
1,050
1,000
950
Macroscopic scale (L0 = 3.97 mm)
900
Mesoscopic scale (L0 = 0.23 mm)
850
800
0 5 · 10−2 0.1 0.15 0.2 0.25 0.3
Longitudinal engineering strain, ε
Fig. 3. Longitudinal response of the specimen at macroscopic and
mesoscopic scales. Net section stress versus engineering strain for
two different gage lengths L0
3 Integrated Digital Image Correlation (I-DIC)
3.1 Principle of I-DIC
Digital Image Correlation (DIC) consists of measuring
displacement fields between a gray level picture f in the
reference configuration and a series of pictures g in the de-
formed configuration [25–27]. The following functional
1 uals χ2c , {Fm } the vector gathering all measured load levels
η2c (t) = (g(x + u(x,t),t) − f (x))2
|Ω| ∑
Ω
is to be minimized with respect to the parametrization of the
measured displacement field u(x,t), where x is any pixel of
the ROI Ω, and t the considered time. Because noise affects
both gray level images f and g, the previous functional is
normalized by 2γ2f
η2c (t)
χ2c (t) = (2)
2γ2f
so that any deviation of correlation residuals from noise in-
duces a level of χ2c (t) that is on average greater than 1 [35],
where γ2f is the variance of image acquisition noise that is
assumed to be white and Gaussian.
The displacement basis for u(x,t) is chosen a priori
to minimize η2c with respect to the associated amplitudes.
Continuous displacement fields found in finite element dis-
cretizations are one possible choice [22, 28, 29], which leads
to lower uncertainties in the measured displacement fields
when compared with local approaches with the same local
displacement interpolations [30]. More importantly, such
fields can be directly compared with finite element simula-
tions. Most of the applications of finite element DIC reported
so far deal with regular meshes made of 4-noded elements.
They are considered herein but with unstructured meshes.
Such type of DIC approach is referred to as Q4-DIC [22,30].
In the following, Integrated DIC will be used to directly
evaluate mechanical fields to extract, say, stress triaxialities.
To have access to such fields, the parameters of the chosen
constitutive law have to be known. They are part of the
proposed procedure, which calibrates them to get the most
reliable estimates of triaxialities for the given constitutive
model. Let {p} denote the vector gathering all unknown ma-
terial parameters. The I-DIC setting consists of minimizing
globally over space and time the total identification residual
0.35 0.4 with respect to the sought material parameters
{popt } = argmin χ2tot ({p}) (3)
{p}
with
1 2 1 2
χ2tot = χ + χ
2 c 2 F
1
χ2c = ∑ χ2c (t) (4)
n n
({Fm } − {Fc ({p})})t ({Fm } − {Fc ({p})})
χ2F ({p}) =
γ2F n
where χ2F denotes the global equilibrium gap, which is to be
minimized in conjunction with the global correlation resid-
(1)
for all considered steps n, {Fc } the vector gathering all com-
puted load levels, and γF is the standard resolution of the load
measurement.
A Newton-Raphson algorithm is implemented to mini-
mize χ2tot , which is nonlinear with respect to {p}. The sen-
sitivities with respect to the sought parameters {p} of the
displacement field u and applied load Fc have to be assessed
iteratively to update the current estimate of the sought pa- will be made hereafter:
rameters
• linear isotropic elasticity,
• elastoplasticity with J2-flow rule and isotropic harden-
u(x,t, {p(i) }) = u(x,t, {p(i−1) }) ing described by Ludwik’s power law [41]
∂u • elastoplasticity with J2-flow rule and isotropic harden-
+ (x,t, {p(i−1) }){δp} (5) ing described by Voce’s exponential law [42]
∂{p}
∂Fc • elastoplasticity with J2-flow rule and isotropic harden-
Fc (t, {p(i) }) = Fc (t, {p(i−1) }) + (t, {p(i−1) }){δp} ing described by extended Voce’s law (i.e., exponential
∂{p}
+ linear hardening).
The identification of Poisson’ ratio ν turned out to be
where {p(i−1) } is the set of parameters at iteration i − 1, and
delicate. The reference value ν = 0.32 was set. In elastic-
{δp} the sought parameter increment. The computation of
ity, only one parameter is sought (i.e., Young’s modulus).
the sensitivity fields ∂u/∂{p} can be performed either ana-
For Ludwik’s law, four parameters are tuned (i.e., Young’s
lytically [33, 40] or numerically by computing the displace-
modulus, yield stress, hardening modulus and hardening ex-
ment fields for small variations of each parameter of the con-
ponent). For Voce’s law, four parameters are also determined
sidered set. In practice, forward finite differences are con-
(i.e., Young’s modulus, yield stress, hardening stress and
sidered to compute the sensitivity fields with the commercial
hardening strain). Last, for extended Voce’s law, five pa-
FE code Abaqus. Similarly, the force sensitivities ∂Fc /∂{p}
rameters are calibrated (i.e., Young’s modulus, yield stress,
are evaluated by resorting to finite forward differences. As
hardening stress, hardening strain, and hardening modulus).
for FEMU, this type of procedure utilizes any FE code, be it
commercial, in a non intrusive way [35].
Both DIC approaches (i.e., Q4-DIC and I-DIC) are to
4 DIC and I-DIC Results
be used in the sequel. Q4-DIC is the reference when the
A first series of DIC analyses is run by resorting to Q4-
same mesh is used in both approaches since no mechanical
DIC in which no mechanical assumptions are made except
hypotheses are made except the continuity of the measured
the fact that the displacement fields are continuous through-
displacement fields. Further, the Dirichlet boundary condi-
out the whole experiment. Figure 4 shows the longitudinal
tions used in I-DIC analyses are those measured via Q4-DIC
displacement field for the last picture prior to failure (i.e.,
so that the comparison of both approaches will be carried
εmacro = 0.072). As expected, there is a very important dis-
out with the same displacements on the loaded boundaries of
placement gradient in the area close to the ligament. The
the ROI. For I-DIC, two different types of mechanical anal-
considered mesh is also shown. It corresponds to the lowest
yses are performed, namely, 2D plane stress simulations and
element size (i.e., 60 pixels) that can be used to reach con-
3D computations. For the latter ones, the Dirichlet bound-
vergence of the Q4-DIC code. Such a coarse mesh is due to
ary conditions correspond to the extruded measured nodal
the poor quality of the random texture. With such conditions,
displacements along the whole thickness of the FE model.
the standard displacement resolution is 0.2 µm (or 0.03 pixel)
Three different meshes are chosen to address numerical con-
and the corresponding strain resolution is 0.05 %. These res-
vergence. First a so-called coarse mesh (i.e., I-DIC-c) is
olutions were determined by analyzing pictures shot for the
identical to that used in Q4-DIC (i.e., 60-pixel elements). An
reference configuration.
intermediate mesh is also constructed and the corresponding
To check the quality of the registration, the dimension-
I-DIC analysis is labelled I-DIC-i (the typical element size is
less residual field ϕ(x,t) = |g(x + u(x,t),t) − f (x)|/∆ f is
equal to 25 pixels). Last an even finer mesh is analyzed (i.e.,
shown in Figure 5 for the final measurement, where ∆ f is
I-DIC-f) with a typical element size of 10 pixels.
the dynamic range of the picture in the reference config-
FE simulations are performed using reduced integration
uration (i.e., ∆ f = maxΩ f − minΩ f ). The mean residual
elements with hourglass stabilization to prevent from volu-
ϕc (t) = hϕ(x,t)iΩ will be shown for each analyzed image
metric locking. It is worth noting that with the present tech-
in Figure 11. The overall mean residual Φc is equal to 0.028,
nique, the mesh can be made as fine as needed since the num-
which indicates that the measured displacements are trust-
ber of unknowns has been significantly reduced to the num-
worthy. However, very high residuals are observed in the
ber of sought material parameters parameters instead of the
ligament region. This is an indication that the chosen kine-
nodal degrees of freedom in Q4-DIC. This in turn provides a
matics associated with the coarse FE mesh is not able to fully
strong robustness of the whole procedure and allows imper-
capture the actual displacement fluctuations. Strain localiza-
fections such as cracking of the paint and poor quality of the
tion may have occurred in that region. Furthermore one ob-
speckle pattern to be dealt with effectively. These challenges
serves paint cracks, represented by large dark spots.
are due to the fact that the sample had to be ground to obtain
From the measured displacement field, it is possible to
the desired thickness.
estimate strain fields. Having a finite element description of
the displacement fields, the displacement gradients are com-
3.2 Constitutive Models puted by differentiating the shape functions as in any stan-
In I-DIC approaches, one important feature is the choice dard mechanical FE code [43]. In the present case the nom-
of the constitutive law. Four different constitutive postulates inal strain component is shown in the longitudinal direction
Fig. 4. Longitudinal displacements (expressed in pixels) measured
Fig. 6. Longitudinal nominal strain field for a Q4-DIC analysis when
via Q4-DIC when εmacro = 0.072
εmacro = 0.072

way since the identified field is mechanically admissible in

0.08
900 an FE sense.
0.07
800

0.06

700

0.05
y (pixels)

600

0.04

500

0.03

400
0.02

300
0.01

200 300 400 500 600 700 800 900 1000 1100
x (pixels)

Fig. 5. Normalized gray level residual field ϕ for a Q4-DIC analysis

when εmacro = 0.072
in Figure 6.
The displacements measured via Q4-DIC on the top and
bottom edges of the ROI are prescribed as Dirichlet bound-
ary conditions to run I-DIC procedures. In the present case,
a 3D mesh and Ludwik’s law are considered. The boundary
conditions are extruded along the thickness direction. Fig-
ure 7 shows the corresponding displacement field measured
via I-DIC-3D-f (i.e., 10-pixel elements are considered). Con-
sequently, a smoother field is measured when compared with
Q4-DIC. Two reasons explain such results. First, the mesh
is finer. Second, I-DIC-3D-f is regularized in a very strong
Fig. 7. Longitudinal displacements (expressed in pixel) measured
via I-DIC-3D-f when εmacro = 0.072
The quality of the registration is again assessed with the
dimensionless residuals for the last analyzed picture (Fig-
ure 8). The mean value over all images and the entire ROI is
equal to 0.033, which indicates that the measured displace-
ments are trustworthy, too. This level is slightly higher than
that observed with Q4-DIC (i.e., 0.028). This is to be ex-
pected when a model error occurs. The latter is mainly lo-
cated in the ligament region. In the other parts of the sample,
the residuals are very low.
 compared with Figures 5 and 8, the levels in the central part
1000 0.08 are significantly lower. These results validate both DIC ap-
proaches and show that kinematic details near the ligament
900 0.07 are captured neither by Q4-DIC nor I-DIC-3D-f for the ana-
lyzed picture.
800 0.06

700 0.05
0.08
y (pixels)

900
600 0.04
0.07
800
500
0.03
0.06

700
400
0.02
0.05

y (pixels)
300 600
0.01
0.04

200 400 600 800 1000 1200 500

x (pixels)
0.03

400
Fig. 8. Normalized gray level residual field ϕ for an I-DIC-3D-f anal- 0.02

ysis when εmacro = 0.072

300
0.01

Figure 9 shows the longitudinal strain field at the last 200 300 400 500 600 700 800 900 1000 1100
step of the experiment. Thanks to I-DIC enabling for finer x (pixels)
meshes, the spatial distribution of strains is better described
in addition to the strain concentrations in the ligament region.
Fig. 10. Absolute difference of normalized gray level residuals be-
tween a Q4-DIC and an I-DIC-3D-f analysis when εmacro = 0.072

Figure 11 shows the change of the mean dimensionless

Fig. 9. Longitudinal nominal strain field for an I-DIC-3D-f analysis
when εmacro = 0.072
To get a more quantitative comparison, Figure 10 reports
the absolute difference of correlation residuals between the
two analyses. Except in the central part of the sample, the
levels are very small and close to noise contributions. When
correlation residuals ϕc for the entire analyzed sequence.
The two DIC analyses follow a very similar trend. When
εmacro ≤ 0.02 the dimensionless residuals are virtually iden-
tical for the two DIC approaches and remain close to the ac-
quisition noise. In this part of the experiment, the chosen
model is fully validated according to the correlation residu-
als. When F ≤ Fmax (i.e., εmacro ≤ 0.034) the dimensionless
residuals are still close for both DIC approaches. The model
is still valid according to the correlation residuals. Con-
versely, when F ≥ Fmax (i.e., εmacro ≥ 0.034) there is a slight
divergence of both residuals. The actual displacement fields
are no longer precisely captured by the coarse DIC mesh.
The quality of the speckle pattern and possibly the cracking
of the paint at the late steps of loading play a major role in
this increase. However from the small difference between
the DIC and I-DIC indicates that the model error remains
acceptable in the high strain regime prior to final failure.
Since global equilibrium was also included in the iden-
tification residual (see Equation (4)), the measured load level
and sum of reaction forces on the edges of the ROI are com-
pared in Figure 12. A very good agreement is observed since
the dimensionless root mean square difference is equal to 11.
Most of the difference occurs in the elastic regime. However,
since the Young’s modulus was one of the parameters to be
optimized and its level is consistent with known values it is
 ·10−2 picture sequence the global identification residual χtot = 9.3.
8
This level is one order of magnitude greater than that ex-
pected if only noise would be detected. A model error is
7
therefore confirmed. In a simpler situation (i.e., strain local-
ization did not occur in a dog-bone sample), levels of the or-
6 der of 5 have been reported [35]. Consequently, even though
a model error is clearly detected, the results should not be
5 discarded.
ϕc

4
18
3
Q4-DIC 16
χtot
I-DIC-3D-f
2 χc
14 χF

1 12

0 10
0 1 2 3 4 5 6 7 8

χf
εmacro ·10−2 8

Fig. 11. Mean dimensionless correlation residuals for all the ana-
6
lyze sequence via Q4-DIC (solid line) and I-DIC-3D-f (dashed line)

4
believed that the discrepancy may be due to experimental
2
imperfections that are not fully captured with 2D-DIC per-
formed on only one face of the sample.
0
0 1 2 3 4 5 6 7 8
εmacro ·10−2
3,000
Fig. 13. Identification, correlation and equilibrium residuals with I-
DIC-3D-f
2,500

To analyze the sensitivity of the present results to nu-

2,000 merical features (i.e., mesh size, 2D vs. 3D simulations) and
model choices, different I-DIC analyses were performed to
assess their influence on the global identification residual χtot
F (N)

1,500 (Table 1). There is a clear gain when 2D plane stress simu-
lations are compared with 3D simulations, for any type of
discretization. Consequently, even though the thickness of
1,000 Exp the sample is very small (i.e., 0.72 mm in the present case),
I-DIC-3D-f a plane stress computation induces a model error that is de-
tected. Concerning the effect of discretization, in the plane
500 stress case the intermediate mesh leads to the lower identifi-
cation residual. The finest 2D mesh leads to very high strain
levels in the ligament, which are unphysical. Conversely,
0
0 1 2 3 4 5 6 7 8
for the 3D simulations, the two finer discretizations lead to
εmacro ·10−2 similar identification residuals. With the chosen constitu-
tive model (i.e., Ludwik’s law), these two discretizations are
Fig. 12. Comparison between measured and identified (via I-DIC- equivalent. When compared with Q4-DIC results, the maxi-
3D-f) load levels. The dash-dotted line corresponds to the maximum mum predicted strains are lower. This result can be explained
load level (Fmax ) by the fact that the paint has cracked (Figure 5) and that Q4-
DIC actually accounts for this spurious effect, which is not
the case of I-DIC that discards it thanks to the elastoplastic
All the previous residuals are now analyzed in terms of regularization.
their distance to noise levels in Figure 13. For the whole The next question is related to the choice of a constitu-
Table 1. Global identification residual χtot for different analyses and 1,250
corresponding maximum triaxiality and cumulated plastic strain

1,200
Approach Constitutive law χtot ηmax pmax
I-DIC-2D-c Ludwik 10.7 0.57 0.39
1,150
I-DIC-2D-i Ludwik 10.5 0.57 0.67
I-DIC-2D-f Ludwik 11.4 0.59 0.91 1,100

σeq (MPa)
I-DIC-3D-c Ludwik 9.7 0.7 0.28
1,050
I-DIC-3D-i Ludwik 9.3 0.69 0.33 Ludwik
Voce
I-DIC-3D-f Ludwik 9.3 0.69 0.36 1,000 ext. Voce

I-DIC-3D-f Voce 9.2 0.69 0.34

I-DIC-3D-f ext. Voce 9.1 0.68 0.34 950

I-DIC-3D-f Elastic 339 0.49 –

900
Q4-DIC – – – 0.43] 0 5 · 10−2 0.1 0.15 0.2 0.25 0.3 0.35 0.4
p
] indicative value (due to paint cracking) Fig. 14. True equivalent stress/plastic strain curves determined for
the three chosen hardening postulates

tive model. To address this issue, the 3D fine mesh is chosen

and the model is changed to a standard and extended version
of Voce’s law. These models induce a second order decrease
of the global identification residual χtot (Table 1). Giving
more freedom in the hardening response of the material does
not induce a significant increase of the identification quality.
The model error is not due to a too poor hardening model.
Conversely, when an elastic law is assumed during the whole
experiment, a very significant degradation of the global iden-
tification residual χtot is observed (Table 1).
In Figure 14, the equivalent stress/plastic strain re-
sponses are shown when described by the three considered
hardening models. The latter ones give very similar predic-
tions for equivalent strain levels less than 0.2. They differ by
about 50 MPa at most beyond that strain level. It is interest-
ing to note that the hardening law with the highest number of
parameters lies mostly between the two hardening postulates
with the lowest degrees of freedom. All these trends confirm
that strain localization has occurred at these high equivalent
strain levels. For all the chosen models, the same level of
static residual is observed. Consequently, all the tested mod-
els are fully compatible in terms of applied load and macro-
scopic strain. Their (small) differences only lie in the dis-
placement fields they induce.
Having no exploitable data with DIC or I-DIC on
smooth samples on the same material tested in tension, the
present results are compared with very recent data in Fig-
ure 15. The calibrated hardening response is compatible with
those provided in Ref. [20] though not identical (i.e., possi-
bly due to differences in the material state). The main differ-
ence lies in the value of the yield stress, which is very sensi-
tive to the initial state of the material. For other grades, the
following range is observed σy = 800 − 1100 MPa % [39],
which encompasses the results of Figure 14. Let us note that
the use of a complex geometry allowed for the calibration
of the hardening curve for much higher levels of equiva-
lent plastic strain than observed in a tensile test (typically
of the order of 0.10-0.16 [20, 39]). Two curves are shown
per analyzed test. The data of Ref. [20] correspond a ten-
sile test on a smooth sample in the LD direction (solid sym-
bols) and the extrapolated model used to predict via FE sim-
ulation the response of an experiment on a notched sample
(solid line). For the present study, the result shown in Fig-
ure 14 for Voce’s model (dashed line) is compared with that
obtained for an identification up to the maximum load level
(F = Fmax ) and subsequently extrapolated for higher strain
levels (thin solid line). The extrapolation underestimates the
actual material response by about 50 MPa.
In the following discussion, the results with Ludwik’s
law are considered for the 3D model with the finest mesh.
This case has led to the smallest identification residual (Ta-
ble 1) for the chosen constitutive equation.
5 Discussion
Having coupled measurement and simulation in a seam-
less (i.e., integrated) way, it is possible to discuss mechan-
ical fields that cannot be directly assessed if full-field mea-
surements were performed independently from simulations.
Since 3D simulations are used to evaluate the sensitivities in
the I-DIC procedure, it is possible to visualize various fields
on the monitored surface but also within the bulk. In the
following discussions, the midsection plane (i.e., parallel to
the observed surface at a distance equal to half the sample
thickness) is also considered.
Figure 16 shows von Mises’ equivalent stress field
σeq (x,t) on the midsection plane for two steps of the loading
history. In elasticity, stress concentrations around the notch
are observed. When plasticity has fully developed, the stress
 1,250

1,200

1,150

1,100
σeq (MPa)

1,050

Voce
1,000 Voce (0 ≤ F ≤ Fmax )
Experiment [20]
Voce [20]
950

900
0 5 · 10−2 0.1 0.15 0.2 0.25 0.3 0.35 0.4 (a) εmacro = 0.004
p

Fig. 15. Comparison of the calibrated hardening curve with the re-
sults of Ref. [20] (thin solid line: extrapolated values). For the present
work, the dashed line corresponds to Voce’s model shown in Fig-
ure 14, the solid line to a calibration up to the maximum load level
(Fmax ) and subsequently extrapolated

concentration has been significantly decreased.

Figure 17 shows von Mises’ equivalent plastic strain
field p(x,t) on the analyzed surface and in the midsection
plane. Even though the overall levels are close, there is a
clear difference between the two fields. In particular, the
field in the midsection plane is more localized than on the
external surface and the maximum level is higher (0.37 vs.
0.29). The fact that the mechanical fields are not identical
on the external surface of the sample when compared with
the midsection is a further confirmation that 3D simulations
are needed, even though they are coupled with 2D surface (b) εmacro = 0.072
measurements.
The triaxiality η(x,t) = σh (x,t)/σeq (x,t) is another field Fig. 16. Von Mises’ equivalent stress maps in the elastic regime (a)
that can be extracted from the FE simulations (Figure 18), and fully developed elastoplastic step (b)
where σh denotes the hydrostatic stress. In the elastic regime,
the field is very similar in terms of distribution in the midsec-
tion plane and on the external surface. The maximum levels free-edge condition. Conversely, the two central points expe-
are very close. rience significant fluctuations of stress triaxiality during the
Conversely, when yielding occurs, the triaxiality field test. The point located in the midsection plane reaches the
is more localized in the midsection plane than on the exter- highest level (i.e., 0.63) at the end of the experiment. A 2D
nal surface (Figure 19). Further, the maximum levels are simulation would not have been able to capture such high
no longer similar (i.e., 0.7 vs. 0.59). These overall trends level.
are similar to those observed for the cumulated plastic strain Table 1 gathers the triaxiality levels for the different
fields shown in Figure 17 for the same loading step. models investigated herein. In elasticity, ηmax = 0.49 for the
Last, the stress triaxiality is shown in Figure 20 for four middle points on the surface and in the midsection plane. A
different points as a function of the corresponding cumulated 2D computation leads to similar levels. Conversely, in plas-
plastic strain. Two of the considered points are located on ticity there is a significant difference between the triaxiality
the notch tip and the other two in the center of the two an- levels observed at the end of the test with 2D simulations
alyzed sections. For the two points close to the notch tip, (i.e., ηmax = 0.57) and 3D results (i.e., ηmax = 0.7 ). This
the stress triaxiality remains virtually constant because of the result further underlines the differences between 2D and 3D
simulations in the present analysis. The hardening law is of
secondary importance and the fact that the global identifi-
cation level was similar led also to very close estimates of
ηmax . Last, in terms of maximum equivalent plastic strain,
the results with I-DIC-3D-f are very close for the three hard-
ening postulates (i.e., pmax ≈ 0.36). When compared with
I-DIC-2D, there is clear difference.

6 Conclusions
It has been shown that triaxiality fields can be estimated
in a thin notched sample made of Ti 6-4 alloy via integrated
DIC. This type of approach seamlessly combines image reg-
istration and finite element simulations. It is worth noting

(a) External surface

(a) External surface

(b) Midsection plane

Fig. 18. Triaxiality maps for two different locations when εmacro =
0.004 (i.e., elastic regime)

that each sample can be analyzed independently and no a pri-

ori information on the material parameters is needed. One of
the additional outputs of such an approach thus is the ma-
terial parameters of the chosen constitutive model, which
are needed to estimate stresses, and the identification qual-
ity based on gray level residuals and global equilibrium gap.
In the studied experiment it was shown that 3D sim-
ulations are needed to lower the correlation residuals and
that fine meshes are desirable to get more trustworthy re-
sults. These meshes are unaccessible with standard DIC
(b) Midsection plane procedures for which the finer the element size the larger
the measurement uncertainties [30]. If the discretization be-
Fig. 17. Von Mises’ equivalent plastic strain maps for two different comes too fine, the registration cannot be performed. To the
locations when εmacro = 0.072 authors’ best of knowledge, it is the first time that I-DIC
results are reported for discretizations that are significantly 0.3
finer than achievable with standard global DIC approaches Face edge
in elastoplastic cases. Further, C8 element have been used Face middle
to properly recover a correct state of pressure around the lig- 0.25 Center edge
ament, a region which is in a nearly incompressible state. Center middle
With I-DIC, the number of unknowns becomes very small
0.2
(i.e., the material parameters), thereby providing a strong
robustness. In the present case, the poor texture and paint
cracking made standard DIC analyses very challenging if not
0.15

p
impossible.
For the notched sample analyzed herein, the maximum
strain levels were at least three times higher than the macro- 0.1
scopic strains to failure evaluated on smooth samples. Thus,
the constitutive laws were probed in a significantly larger
5 · 10−2

0
0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65
η

Fig. 20. Triaxiality history for four different points either located at
the notch tip or in the center of the considered surfaces (i.e., external
plane or midsection plane)

strain range. In terms of identification levels, the three inves-

tigated isotropic hardening laws led to similar global iden-
tification residuals. The detail of the chosen model seems
secondary except for elasticity that led to very large errors.
However, beyond the ultimate load level, significant errors
detected on the gray level residuals indicate model errors as-
sociated with strain localization. Conversely, the load resid-
uals remained of the same order of magnitude as in the first
part of the experiment.
It is shown that the estimation of triaxiality fields re-
(a) External surface quires 3D simulations even when the considered samples are
thin. This is especially true when elastoplasticity is fully de-
veloped. The maximum levels of triaxiality are reached in
the bulk of the sample (i.e., What is essential is invisible to
the eye1 ). In the present case, the maximum triaxiality is vir-
tually independent of the details of the isotropic hardening
law. However, it is significantly different from the levels ob-
served in elasticity or even with 2D plane stress simulations.
The next step of such analyses is to enrich the consti-
tutive law to lower the identification residuals. In Ref. [20]
anisotropy of the plastic response was observed and mod-
eled. Similarly, twinning may occur. The I-DIC procedure
developed herein can be extended to anisotropic yield sur-
faces and more complex hardening models. A damage model
may also be coupled with the hardening law. In such consti-
tutive models, the role of stress triaxiality is known to be of
primary importance [6, 15–18].
Last, only one external surface was observed and the
measured (Dirichlet) boundary conditions were extruded for
3D simulations. This hypothesis may induce errors that can
(b) Midsection plane

Fig. 19. Triaxiality maps for two different locations when εmacro =
0.072 (i.e., plastic regime) 1 Antoine de Saint-Exupéry (1943) “The Little Prince” Reynal & Hitch-

cock (USA)
be lowered by using at least another camera (for the oppo-
site surface) or by applying a regularization technique on
the boundary conditions [44, 45]. If the sample thickness in-
creases, four instead of two cameras may be utilized. Stere-
ocorrelation may also be considered to evaluate 3D surface
displacements. Another route is also possible by resorting to
computed tomography [46] or laminography [47] in which
digital volume correlation is a useful tool to reveal strain lo-
calization induced by plasticity [48]. All these experimen-
tal improvements may add more information that is useful
when, say, integrated approaches are to be implemented.
Acknowledgements
The experiment reported herein was performed with the
help of Xavier Pinelli. The I-DIC code used herein is part
of the Metil platform developed by Hugo Leclerc [49]. This
work has been funded by Turbomeca (SAFRAN Group). It
is also a pleasure to acknowledge the support of Région Ile
de France (“DICCIT” project).
References
[1] Robinson, E., 1944. “Bursting tests of steam-turbine
disk wheels”. Trans. ASME, 66, pp. 373–386.
[2] Mazière, M., Besson, J., Forest, S., Tanguy, B.,
Chalons, H., and Vogel, F., 2009. “Overspeed burst
of elastoviscoplastic rotating disks – Part I: Analytical
and numerical stability analyses”. Eur. J. Mechanics -
A/Solids, 28(1), pp. 36–44.
[3] Mazière, M., Besson, J., Forest, S., Tanguy, B.,
Chalons, H., and Vogel, F., 2009. “Overspeed burst
of elastoviscoplastic rotating disks – Part II: Burst of a
superalloy turbine disk”. Eur. J. Mechanics - A/Solids,
28(3), pp. 428–432.
[4] Hosford, W., 1972. “A generalized isotropic yield cri-
terion”. J. Appl. Mech., 39, pp. 607–609.
[5] McClintock, F., 1968. “A criterion for ductile fracture
by growth of holes”. J. Appl. Mech., 35(2), pp. 363–
371.
[6] Rice, J., and Tracey, D., 1969. “On the ductile enlarge-
ment of voids in triaxial stress fields”. J. Mech. Phys.
Solids, 17, pp. 201–217.
[7] Lode, W., 1925. “The influence of the intermediate
principal stress on yielding and failure of iron, copper
and nickel”. Zeits. Angew. Math. Mech., 5, pp. 142–
149.
[8] Zhang, K., Bai, J., and François, D., 2001. “Numerical
analysis of the influence of the lode parameter on void
growth”. Int. J. Solids Struct., 38(32-33), pp. 5847 –
5856.
[9] Bao, Y., and Wierzbicki, T., 2004. “On fracture locus
in the equivalent strain and stress triaxiality space”. Int.
J. Mech. Sci, 46(1), pp. 81–98.
[10] Barsoum, I., and Faleskog, J., 2007. “Rup-
ture mechanisms in combined tension and shear-
micromechanics”. Int. J. Solids Struct., 44(17),
pp. 5481–5498.
[11] Nahshon, K., and Hutchinson, J., 2008. “Modification
of the gurson model for shear failure”. Eur. J. Mech.
A/Solids, 27, pp. 1–17.
[12] Dunand, M., and Mohr, D., 2011. “On the predictive
capabilities of the shear modified gurson and the mod-
ified mohr coulomb fracture models over a wide range
of stress triaxialities and lode angles”. J. Mech. Phys.
Solids, 59(7), pp. 1374–1394.
[13] Gao, X., Zhang, T., Zhou, J., Graham, S., Hayden, M.,
and Roe, C., 2011. “On stress-state dependent plastic-
ity modeling: Significance of the hydrostatic stress, the
third invariant of stress deviator and the non-associated
flow rule”. Int. J. Plasticity, 27(2), pp. 217–231.
[14] Malcher, L., Andrade Pires, F., and César de Sá, J.,
2012. “An assessment of isotropic constitutive models
for ductile fracture under high and low stress triaxial-
ity”. Int. J. Plast., 30-31, pp. 81–115.
[15] Gurson, A., 1977. “Continuum theory of ductile rup-
ture by void nucleation and growth: Part i - yield crite-
rion and flow rules for porous ductile media”. ASME J.
Eng. Mat. Techn., 99, pp. 2–15.
[16] Lemaitre, J., 1984. “How to use damage mechanics”.
Nucl. Eng. Design, 80, pp. 233–245.
[17] Needleman, A., and Tvergaard, V., 1984. “An analy-
sis of ductile rupture in notched bars”. J. Mech. Phys.
Solids, 32(6), pp. 461–490.
[18] Tvergaard, V., and Needleman, A., 1984. “Analysis
of the cup-cone fracture in a round tensile bar”. Acta
Metall., 32(1), pp. 157–169.
[19] Koplik, J., and Needleman, A., 1988. “Void growth
and coalescence in porous plastic solids”. Int. J. Solids
Struct., 24(8), pp. 835–853.
[20] Tuninetti, V., Gilles, G., Milis, O., Pardoen, T.,
and Habraken, A., 2015. “Anisotropy and tension-
compression asymmetry modeling of the room temper-
ature plastic response of ti-6al-4v”. Int. J. Plasticity,
67, pp. 53–68.
[21] Wattrisse, B., Chrysochoos, A., Muracciole, J., and
Némoz-Gaillard, M., 2001. “Analysis of strain local-
isation during tensile test by digital image correlation”.
Exp. Mech., 41(1), pp. 29–39.
[22] Besnard, G., Hild, F., and Roux, S., 2006. ““finite-
element” displacement fields analysis from digital im-
ages: Application to portevin-le châtelier bands”. Exp.
Mech., 46, pp. 789–803.
[23] Rastogi, P., ed., 2000. Photomechanics, Vol. 77 of Top-
ics in Applied Physics. Springer, Berlin (Germany).
[24] Rastogi, P., and Hack, E., eds., 2012. Optical Methods
for Solid Mechanics. A Full-Field Approach. Wiley-
VCH, Berlin.
[25] Hild, F., and Roux, S., 2012. Digital Image Correla-
tion. Wiley-VCH, Weinheim (Germany), pp. 183–228.
[26] Sutton, M., 2013. “Computer vision-based, non-
contacting deformation measurements in mechanics:
A generational transformation”. Appl. Mech. Rev.,
65(AMR-13-1009, 050802).
[27] Sutton, M., Orteu, J., and Schreier, H., 2009. Image
correlation for shape, motion and deformation mea-
 surements: Basic Concepts, Theory and Applications. ments”. Eng. Comput., 22(5-6), pp. 487–504.
Springer, New York, NY (USA). [45] Feissel, P., and Allix, O., 2007. “Modified constitutive
[28] Broggiato, G., 2004. “Adaptive image correlation tech- relation error identification strategy for transient dy-
nique for full-field strain measurement”. In 12th Int. namics with corrupted data: The elastic case”. Comput.
Conf. Exp. Mech., C. Pappalettere, ed., McGraw Hill, Meth. Appl. Mech. Eng., 196(13/16), pp. 1968–1983.
Lilan (Italy), pp. 420–421. [46] Maire, E., and Withers, P. J., 2014. “Quantitative x-ray
[29] Sun, Y., Pang, J., Wong, C., and Su, F., 2005. “Finite- tomography”. Int. Mat. Rev., 59(1), pp. 1–43.
element formulation for a digital image correlation [47] Morgeneyer, T., Helfen, L., Sinclair, I., Proudhon, H.,
method”. Appl. Optics, 44(34), pp. 7357–7363. Xu, F., and Baumbach, T., 2011. “Ductile crack initi-
[30] Hild, F., and Roux, S., 2012. “Comparison of local and ation and propagation assessed via in situ synchrotron
global approaches to digital image correlation”. Exp. radiation computed laminography”. Scripta Mat., 65,
Mech., 52(9), pp. 1503–1519. pp. 1010–1013.
[31] Hild, F., and Roux, S., 2006. “Digital image corre- [48] Morgeneyer, T., Taillandier-Thomas, T., Helfen, L.,
lation: From measurement to identification of elastic Baumbach, T., Sinclair, I., Roux, S., and Hild, F., 2014.
properties - a review”. Strain, 42, pp. 69–80. “In situ 3d observation of early strain localisation dur-
[32] Roux, S., and Hild, F., 2006. “Stress intensity fac- ing failure of thin al alloy (2198) sheet”. Acta Mat., 69,
tor measurements from digital image correlation: post- pp. 78–91.
processing and integrated approaches”. Int. J. Fract., [49] Leclerc, H., 2007. “Plateforme metil : optimisations et
140(1-4), pp. 141–157. facilités liées à la génération de code”. In 8e Colloque
[33] Leclerc, H., Périé, J., Roux, S., and Hild, F., 2009. In- National en Calcul des Structures.
tegrated digital image correlation for the identification
of mechanical properties, Vol. LNCS 5496. Springer,
Berlin, pp. 161–171.
[34] Réthoré, J., 2010. “A fully integrated noise robust strat-
egy for the identification of constitutive laws from dig-
ital images”. Int. J. Num. Meth. Eng., 84(6), pp. 631–
660.
[35] Mathieu, F., Leclerc, H., Hild, F., and Roux, S.,
2015. “Estimation of elastoplastic parameters via
weighted FEMU and Integrated-DIC”. Exp. Mech.,
55(1), pp. 105–119.
[36] Kavanagh, K., and Clough, R., 1971. “Finite element
applications in the characterization of elastic solids”.
Int. J. Solids Struct., 7, pp. 11–23.
[37] Grédiac, M., and Hild, F., eds., 2012. Full-Field Mea-
surements and Identification in Solid Mechanics. ISTE
/ Wiley, London (UK).
[38] Ghahremaninezhad, A., and Ravi-Chandar, K., 2012.
“Ductile failure behavior of polycrystalline al 6061-t6”.
Int. J. Fract., 174(2), pp. 177–202.
[39] Abouridouane, M., 2005. “Bruchverhalten von leicht-
metallen unter impact-beanspruchung” (in German).
Phd thesis.
[40] Cooreman, S., Lecompte, D., Sol, H., Vantomme, J.,
and Debruyne, D., 2007. “Elasto-plastic material pa-
rameter identification by inverse methods: Calculation
of the sensitivity matrix”. Int. J. Solids Struct., 44(13),
pp. 4329–4341.
[41] Ludwik, P., 1909. Elemente der technologischen
Mechanik. Verlag Von Julius Springer (Leipzig, Ger-
many).
[42] Voce, E., 1948. “The relationship between stress and
strain for homogeneous deformation”. J. Inst. Met., 74,
pp. 537–562.
[43] Zienkievicz, O., and Taylor, R., 1989. The Finite Ele-
ment Method. 4th edition. McGraw-Hill, London (UK).
[44] Allix, O., Feissel, P., and Nguyen, H., 2005. “Identi-
fication strategy in the presence of corrupted measure-